--- title: Title keywords: fastai sidebar: home_sidebar nb_path: "05b05 Test Swiss Roll with New Kernel (7-11).ipynb" ---
{% raw %}
{% endraw %} {% raw %} 
from directed_graphs.datasets import directed_circle, directed_cylinder, directed_spiral, directed_swiss_roll, directed_spiral_uniform, directed_swiss_roll_uniform
from directed_graphs.datasets import plot_directed_2d, plot_directed_3d
from directed_graphs.diffusion_flow_embedding import DiffusionFlowEmbedder
import torch
import numpy as np
{% endraw %} {% raw %} 
if torch.__version__[:4] == "1.13":
	# device = torch.device('cuda' if torch.cuda.is_available() else 'mps' if torch.has_mps else 'cpu')
	device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

else:
	device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
device

device(type='cuda')
{% endraw %} {% raw %} 
X, flow, labels = directed_swiss_roll_uniform(num_nodes=1000, num_spirals=2.5, radius=1, height=20, xtilt=0, ytilt=0)
plot_directed_3d(X, flow, labels, mask_prob=0.5)
{% endraw %} {% raw %} 
X = torch.tensor(X)
flow = torch.tensor(flow)
X = X.float().to(device)
flow = flow.float().to(device)
{% endraw %} {% raw %} 
dfe = DiffusionFlowEmbedder(X,flow,t=1,sigma_graph=5,sigma_embedding=5,device=device, learning_rate = 1e-3)
dfe = dfe.to(device)
embeddings = dfe.fit(n_steps=2000)

  0%|          | 0/2000 [00:00<?, ?it/s]
EPOCH 0. Loss 1450.9156494140625. Flow strength 4.999000072479248. Weight of flow 0.5 Heatmap of P embedding is 

  5%|▌         | 100/2000 [00:30<09:17,  3.41it/s]
EPOCH 100. Loss 1838.3935546875. Flow strength 4.937063217163086. Weight of flow 0.5 Heatmap of P embedding is 

 10%|â–ˆ         | 200/2000 [00:55<06:41,  4.48it/s]
EPOCH 200. Loss 452.94805908203125. Flow strength 4.917881965637207. Weight of flow 0.5 Heatmap of P embedding is 

 15%|█▌        | 300/2000 [01:18<06:17,  4.51it/s]
EPOCH 300. Loss 35.4312744140625. Flow strength 4.897221088409424. Weight of flow 0.5 Heatmap of P embedding is 

 20%|██        | 400/2000 [01:42<05:56,  4.49it/s]
EPOCH 400. Loss 12.34261703491211. Flow strength 4.875749588012695. Weight of flow 0.5 Heatmap of P embedding is 

 25%|██▌       | 500/2000 [02:05<05:34,  4.49it/s]
EPOCH 500. Loss 8.347378730773926. Flow strength 4.861460208892822. Weight of flow 0.5 Heatmap of P embedding is 

 30%|███       | 600/2000 [02:29<05:09,  4.52it/s]
EPOCH 600. Loss -0.7507753372192383. Flow strength 4.843987464904785. Weight of flow 0.5 Heatmap of P embedding is 

 35%|███▌      | 700/2000 [02:55<06:20,  3.41it/s]
EPOCH 700. Loss -0.3254380226135254. Flow strength 4.823906421661377. Weight of flow 0.5 Heatmap of P embedding is 

 40%|████      | 800/2000 [03:20<04:25,  4.52it/s]
EPOCH 800. Loss 2.6325457096099854. Flow strength 4.8075642585754395. Weight of flow 0.5 Heatmap of P embedding is 

 45%|████▌     | 900/2000 [03:44<04:06,  4.47it/s]
EPOCH 900. Loss 1.7728157043457031. Flow strength 4.785553455352783. Weight of flow 0.5 Heatmap of P embedding is 

 50%|█████     | 1000/2000 [04:07<03:44,  4.45it/s]
EPOCH 1000. Loss 1.8203563690185547. Flow strength 4.75385856628418. Weight of flow 0.5 Heatmap of P embedding is 

 55%|█████▌    | 1100/2000 [04:32<04:23,  3.41it/s]
EPOCH 1100. Loss 0.15930986404418945. Flow strength 4.719078063964844. Weight of flow 0.5 Heatmap of P embedding is 

 60%|██████    | 1200/2000 [05:00<02:57,  4.51it/s]
EPOCH 1200. Loss -0.749056339263916. Flow strength 4.671835899353027. Weight of flow 0.5 Heatmap of P embedding is 

 65%|██████▌   | 1300/2000 [05:29<03:25,  3.41it/s]
EPOCH 1300. Loss 0.5551466941833496. Flow strength 4.615596771240234. Weight of flow 0.5 Heatmap of P embedding is 

 70%|███████   | 1400/2000 [05:54<02:13,  4.49it/s]
EPOCH 1400. Loss 0.5920677185058594. Flow strength 4.564528465270996. Weight of flow 0.5 Heatmap of P embedding is 

 75%|███████▌  | 1500/2000 [06:18<01:51,  4.50it/s]
EPOCH 1500. Loss 5.697395324707031. Flow strength 4.495549201965332. Weight of flow 0.5 Heatmap of P embedding is 

 80%|████████  | 1600/2000 [06:42<01:28,  4.50it/s]
EPOCH 1600. Loss 2.416745185852051. Flow strength 4.422905921936035. Weight of flow 0.5 Heatmap of P embedding is 

 85%|████████▌ | 1700/2000 [07:09<01:27,  3.41it/s]
EPOCH 1700. Loss 3.153214931488037. Flow strength 4.349822998046875. Weight of flow 0.5 Heatmap of P embedding is 

 90%|█████████ | 1800/2000 [07:40<00:58,  3.40it/s]
EPOCH 1800. Loss 3.8875269889831543. Flow strength 4.288753509521484. Weight of flow 0.5 Heatmap of P embedding is 

 95%|█████████▌| 1900/2000 [08:10<00:29,  3.41it/s]
EPOCH 1900. Loss 3.8878660202026367. Flow strength 4.22802209854126. Weight of flow 0.5 Heatmap of P embedding is 

100%|██████████| 2000/2000 [08:39<00:00,  3.85it/s]

Exiting training with loss  tensor(5.1083, device='cuda:0', grad_fn=<AddBackward0>)
{% endraw %} {% raw %} 
dfe.visualize_points(labels)

No artists with labels found to put in legend.  Note that artists whose label start with an underscore are ignored when legend() is called with no argument.
{% endraw %} {% raw %} 
dfe2 = DiffusionFlowEmbedder(X,flow,t=1,
                             sigma_graph=5,sigma_embedding=5,device=device, 
                             learning_rate = 1e-3, smoothness = 0, weight_of_flow = 1)
dfe2 = dfe2.to(device)
embeddings = dfe2.fit(n_steps=2000)

  0%|          | 0/2000 [00:00<?, ?it/s]
EPOCH 0. Loss 0.7735907435417175. Flow strength 4.999000072479248. Weight of flow 1 Heatmap of P embedding is 

  5%|▌         | 100/2000 [00:26<07:08,  4.44it/s]
EPOCH 100. Loss 0.7301082015037537. Flow strength 4.908123970031738. Weight of flow 1 Heatmap of P embedding is 

 10%|â–ˆ         | 200/2000 [00:49<06:46,  4.43it/s]
EPOCH 200. Loss 0.7031649947166443. Flow strength 4.836468696594238. Weight of flow 1 Heatmap of P embedding is 

 15%|█▌        | 300/2000 [01:13<06:20,  4.47it/s]
EPOCH 300. Loss 0.7496340274810791. Flow strength 4.802023410797119. Weight of flow 1 Heatmap of P embedding is 

 20%|██        | 400/2000 [01:40<07:56,  3.35it/s]
EPOCH 400. Loss 0.7198862433433533. Flow strength 4.740354061126709. Weight of flow 1 Heatmap of P embedding is 

 25%|██▌       | 500/2000 [02:11<07:25,  3.37it/s]
EPOCH 500. Loss 0.6321343183517456. Flow strength 4.8584771156311035. Weight of flow 1 Heatmap of P embedding is 

 30%|███       | 600/2000 [02:37<05:13,  4.46it/s]
EPOCH 600. Loss 0.6395624279975891. Flow strength 4.955665588378906. Weight of flow 1 Heatmap of P embedding is 

 35%|███▌      | 700/2000 [03:01<04:54,  4.42it/s]
EPOCH 700. Loss 0.6584041118621826. Flow strength 4.963932991027832. Weight of flow 1 Heatmap of P embedding is 

 40%|████      | 800/2000 [03:25<04:29,  4.46it/s]
EPOCH 800. Loss 0.690463662147522. Flow strength 4.952167987823486. Weight of flow 1 Heatmap of P embedding is 

 45%|████▌     | 900/2000 [03:48<04:14,  4.31it/s]
EPOCH 900. Loss 0.697661280632019. Flow strength 5.0122551918029785. Weight of flow 1 Heatmap of P embedding is 

 50%|█████     | 1000/2000 [04:12<03:43,  4.48it/s]
EPOCH 1000. Loss 0.6644861102104187. Flow strength 4.899844646453857. Weight of flow 1 Heatmap of P embedding is 

 55%|█████▌    | 1100/2000 [04:36<03:21,  4.47it/s]
EPOCH 1100. Loss 0.6746009588241577. Flow strength 4.959043025970459. Weight of flow 1 Heatmap of P embedding is 

 60%|██████    | 1200/2000 [05:00<02:59,  4.47it/s]
EPOCH 1200. Loss 0.7039652466773987. Flow strength 4.906436920166016. Weight of flow 1 Heatmap of P embedding is 

 65%|██████▌   | 1300/2000 [05:23<02:36,  4.48it/s]
EPOCH 1300. Loss 0.7418500185012817. Flow strength 4.879170894622803. Weight of flow 1 Heatmap of P embedding is 

 70%|███████   | 1400/2000 [05:48<02:15,  4.42it/s]
EPOCH 1400. Loss 0.7125900983810425. Flow strength 4.840127944946289. Weight of flow 1 Heatmap of P embedding is 

 75%|███████▌  | 1500/2000 [06:12<01:52,  4.44it/s]
EPOCH 1500. Loss 0.6996027231216431. Flow strength 4.840446472167969. Weight of flow 1 Heatmap of P embedding is 

 80%|████████  | 1600/2000 [06:38<01:37,  4.09it/s]
EPOCH 1600. Loss 0.6573364734649658. Flow strength 4.842692852020264. Weight of flow 1 Heatmap of P embedding is 

 85%|████████▌ | 1700/2000 [07:03<01:28,  3.38it/s]
EPOCH 1700. Loss 0.6391164660453796. Flow strength 4.839215278625488. Weight of flow 1 Heatmap of P embedding is 

 90%|█████████ | 1800/2000 [07:28<00:44,  4.47it/s]
EPOCH 1800. Loss 0.5090253353118896. Flow strength 4.84824275970459. Weight of flow 1 Heatmap of P embedding is 

 95%|█████████▌| 1900/2000 [07:54<00:29,  3.39it/s]
EPOCH 1900. Loss 0.44025570154190063. Flow strength 4.803142070770264. Weight of flow 1 Heatmap of P embedding is 

100%|██████████| 2000/2000 [08:18<00:00,  4.01it/s]
Exiting training with loss  tensor(0.4159, device='cuda:0', grad_fn=<AddBackward0>)














    {% endraw %}

    {% raw %}
    







    


dfe2.visualize_points(labels)



    

















No artists with labels found to put in legend.  Note that artists whose label start with an underscore are ignored when legend() is called with no argument.




























    {% endraw %}

    {% raw %}
    







    


X, flow, labels = directed_swiss_roll_uniform(num_nodes=1000, num_spirals=2.5, radius=5, height=5, xtilt=0, ytilt=0)
plot_directed_3d(X, flow, labels, mask_prob=0.5)

X = torch.tensor(X)
flow = torch.tensor(flow)
X = X.float().to(device)
flow = flow.float().to(device)



    
































    {% endraw %}

    {% raw %}
    







    


dfe3 = DiffusionFlowEmbedder(X,flow,t=1,
                             sigma_graph=10,sigma_embedding=10,device=device, 
                             learning_rate = 1e-3, smoothness = 0, weight_of_flow = 1)
dfe3 = dfe3.to(device)
embeddings = dfe3.fit(n_steps=2000)



    

















  0%|          | 0/2000 [00:00<?, ?it/s]












EPOCH 0. Loss 1.7039836645126343. Flow strength 4.999000072479248. Weight of flow 1 Heatmap of P embedding is 









































  5%|▌         | 100/2000 [00:23<07:06,  4.46it/s]












EPOCH 100. Loss 1.1169966459274292. Flow strength 5.062305450439453. Weight of flow 1 Heatmap of P embedding is 









































 10%|â–ˆ         | 200/2000 [00:48<06:43,  4.47it/s]












EPOCH 200. Loss 0.7491041421890259. Flow strength 5.149476051330566. Weight of flow 1 Heatmap of P embedding is 









































 15%|█▌        | 300/2000 [01:15<08:21,  3.39it/s]












EPOCH 300. Loss 0.5054308176040649. Flow strength 5.180909633636475. Weight of flow 1 Heatmap of P embedding is 









































 20%|██        | 400/2000 [01:39<06:01,  4.43it/s]












EPOCH 400. Loss 0.2589603066444397. Flow strength 5.179901123046875. Weight of flow 1 Heatmap of P embedding is 









































 25%|██▌       | 500/2000 [02:03<05:35,  4.48it/s]












EPOCH 500. Loss 0.18674583733081818. Flow strength 5.1641011238098145. Weight of flow 1 Heatmap of P embedding is 









































 30%|███       | 600/2000 [02:26<05:13,  4.46it/s]












EPOCH 600. Loss 0.12053244560956955. Flow strength 5.161307334899902. Weight of flow 1 Heatmap of P embedding is 









































 35%|███▌      | 700/2000 [02:50<04:50,  4.48it/s]












EPOCH 700. Loss 0.0954655334353447. Flow strength 5.16131591796875. Weight of flow 1 Heatmap of P embedding is 









































 40%|████      | 800/2000 [03:13<04:28,  4.48it/s]












EPOCH 800. Loss 0.0893363207578659. Flow strength 5.161392688751221. Weight of flow 1 Heatmap of P embedding is 









































 45%|████▌     | 900/2000 [03:37<04:06,  4.47it/s]












EPOCH 900. Loss 0.08691702783107758. Flow strength 5.161564826965332. Weight of flow 1 Heatmap of P embedding is 









































 50%|█████     | 1000/2000 [04:01<03:43,  4.48it/s]












EPOCH 1000. Loss 0.08531311899423599. Flow strength 5.161726474761963. Weight of flow 1 Heatmap of P embedding is 









































 55%|█████▌    | 1100/2000 [04:24<03:21,  4.47it/s]












EPOCH 1100. Loss 0.0847654938697815. Flow strength 5.161852836608887. Weight of flow 1 Heatmap of P embedding is 









































 60%|██████    | 1200/2000 [04:54<03:56,  3.39it/s]












EPOCH 1200. Loss 0.0844942256808281. Flow strength 5.161948204040527. Weight of flow 1 Heatmap of P embedding is 









































 65%|██████▌   | 1300/2000 [05:25<03:26,  3.39it/s]












EPOCH 1300. Loss 0.08430938422679901. Flow strength 5.162043571472168. Weight of flow 1 Heatmap of P embedding is 









































 70%|███████   | 1400/2000 [05:56<02:56,  3.39it/s]












EPOCH 1400. Loss 0.08415163308382034. Flow strength 5.162138938903809. Weight of flow 1 Heatmap of P embedding is 









































 75%|███████▌  | 1500/2000 [06:26<02:27,  3.39it/s]












EPOCH 1500. Loss 0.08401276916265488. Flow strength 5.1621904373168945. Weight of flow 1 Heatmap of P embedding is 









































 80%|████████  | 1600/2000 [06:56<01:29,  4.47it/s]












EPOCH 1600. Loss 0.0839201956987381. Flow strength 5.162238121032715. Weight of flow 1 Heatmap of P embedding is 









































 85%|████████▌ | 1700/2000 [07:20<01:06,  4.50it/s]












EPOCH 1700. Loss 0.08383737504482269. Flow strength 5.162285804748535. Weight of flow 1 Heatmap of P embedding is 









































 90%|█████████ | 1800/2000 [07:44<00:48,  4.11it/s]












EPOCH 1800. Loss 0.08377061784267426. Flow strength 5.1623334884643555. Weight of flow 1 Heatmap of P embedding is 









































 95%|█████████▌| 1900/2000 [08:10<00:22,  4.49it/s]












EPOCH 1900. Loss 0.08369430154561996. Flow strength 5.162381172180176. Weight of flow 1 Heatmap of P embedding is 









































100%|██████████| 2000/2000 [08:34<00:00,  3.89it/s]












Exiting training with loss  tensor(0.0836, device='cuda:0', grad_fn=<AddBackward0>)


























    {% endraw %}

    {% raw %}
    







    


dfe3.visualize_points(labels)



    

















No artists with labels found to put in legend.  Note that artists whose label start with an underscore are ignored when legend() is called with no argument.




























    {% endraw %}

    {% raw %}
    







    


import phate
import matplotlib.pyplot as plt



    








    {% endraw %}

    {% raw %}
    







    


X, flow, labels = directed_swiss_roll_uniform(num_nodes=1000, num_spirals=2.5, radius=5, height=5, xtilt=0, ytilt=0)
plot_directed_3d(X, flow, labels, mask_prob=0.5)



    
































    {% endraw %}

    {% raw %}
    







    


phate_operator = phate.PHATE(n_components = 2, knn=5, decay=20, t=150)
phate_op = phate.PHATE()
data_phate = phate_op.fit_transform(X)



    

















Calculating PHATE...
  Running PHATE on 1000 observations and 3 variables.
  Calculating graph and diffusion operator...
    Calculating KNN search...
    Calculated KNN search in 0.02 seconds.
    Calculating affinities...
    Calculated affinities in 0.02 seconds.
  Calculated graph and diffusion operator in 0.05 seconds.
  Calculating optimal t...
    Automatically selected t = 20
  Calculated optimal t in 0.28 seconds.
  Calculating diffusion potential...
  Calculated diffusion potential in 0.09 seconds.
  Calculating metric MDS...
  Calculated metric MDS in 1.96 seconds.
Calculated PHATE in 2.39 seconds.














    {% endraw %}

    {% raw %}
    







    


phate.plot.scatter2d(data_phate, c=labels)



    

















/tmp/ipykernel_249558/1966985081.py:1: DeprecationWarning: Call to deprecated function (or staticmethod) scatter2d. (Use scprep.plot.scatter2d instead) -- Deprecated since version 1.0.0.
  phate.plot.scatter2d(data_phate, c=labels)















<AxesSubplot:>




























    {% endraw %}